Keep URLs as Scalar Metadata, Not Predicates

URLs preserving the external location of a source appear as scalar body content — a URL: line in the body (preferred for a single canonical URL) or bulleted entries inside a ## Sources section (when multiple URLs are material). URLs MUST NOT be encoded as named-edge predicate values: no external_url::https://..., no source_location::[[https://...]]↗ wrapping a URL as a pseudo-wikilink target. Graph-structural claims about the source (what role it plays, what it informs, what supersedes it) live in named edges; the URL lives in body scalar form.

Why

The commitment preserves graph spine clean of flat data. Named-edge predicates carry graph-structural semantics — they connect nodes by typed relations that the graph's reader traverses for meaning. URLs connect nodes to external resources, but the connection's shape is different: a URL is a flat pointer to a document, not a typed relation between two graph vertices. Encoding URLs as predicate values would either misuse the predicate vocabulary (predicates become dual-purpose — some carry structural claims, others carry flat data) or force wiki-style wrapping of URLs as pseudo-targets ([[https://example.com]]↗), which the wiki layer does not resolve as a node.

The separation makes structural and flat data legible at the tier layer. A reader scanning a Reference's body for the external location finds the URL: line or the ## Sources section — predictable placements that carry URLs specifically. A reader traversing the Reference's graph relations finds serves_as::, informs::, supersedes:: — predicate edges that carry structural claims. The two readers have different jobs; keeping URLs in scalar form and structural claims in predicate form lets each reader find what they need at the tier the prototype's vocabulary reserves for it.

The rule applies specifically to URLs-as-artifact-locations. A Reference's canonical URL (where the artifact lives) is scalar. A typed edge to another graph node by wikilink is not — serves_as::[[Primary Convention Source]] is a named edge whose target is a graph node, not a URL. The rule distinguishes URLs (flat, scalar, body content) from wikilinks (typed, graph-participating, named edges). Wiki tooling's ↗ marker applies to external-tradition wikilink targets (nodes in other wikis), not to URLs — the marker signals "this target lives elsewhere in wiki space," not "this is a URL."

The prototype's ## Sources convention handles multi-URL cases. When a Reference has a primary URL plus a mirror, a running instance, or a supplementary file, the ## Sources section lists each with a brief annotation. The section is scalar body content; its entries are not typed edges. If a source becomes load-bearing enough to warrant its own Reference node, it is promoted to a stub node under sources/ (a future folder), and the Reference cites the stub by wikilink via a typed predicate — which re-enters graph participation appropriately.

Alternatives Considered

external_url::[[https://...]] predicate. Wrap URLs as pseudo-wikilink targets on a predicate edge. Rejected because the wrapping breaks wikilink resolution. [[https://example.com]] resolves by filename — a file named https://example.com.md — not by URL fetch. The predicate becomes a named-edge that points at a ghost link for every external URL, polluting the ghost-link signal (which should mean "intended future node") with "this is actually a URL, not a node." The wikilink convention is wrong for URLs.

Predicate URLs without wikilink wrapping. Let URLs appear as bare predicate values: external_url::https://example.com. Rejected because predicates are the graph's typed-edge syntax; non-wikilink values violate the convention that predicates target nodes. Mixing URL values and wikilink values in the predicate vocabulary would produce traversal ambiguity — a reader filtering on external_url:: would need to check whether each value was a URL or a wikilink.

YAML-encoded URLs. Carry URLs in YAML frontmatter (url: https://...). Rejected by [[Restrict YAML to Scalar Metadata]] for named-edge predicates, but URLs-as-scalars could live in YAML. The body placement is preferred because References' URLs are load-bearing reading content — a reader opening the Reference expects the URL visible in the body near the top, not buried in frontmatter. The rule does not strictly forbid YAML URLs; it establishes body placement as canonical.

What Would Change It

The commitment would be revisited under one condition.

URL-as-predicate proves valuable for traversal queries. If a compelling query use case emerges — traverse the graph, collect all external URLs for citation, filter by URL pattern — and the scalar body placement makes the query awkward, the convention could weaken to permit URL-predicates. The alternative is a build step that extracts URLs from body URL: lines into a query index, which preserves the scalar-body convention while enabling the traversal. Current use patterns have not surfaced the need; the revisit condition is latent.

Relations